OMB M-25-21 / M-25-22 — Federal AI Use and Procurement
Summary
Section titled “Summary”Two companion OMB memoranda, issued by the Office of Management and Budget on 2025-04-03. Paired with EO 14179 (2025-01), they build the foundation of federal government AI governance under the Trump administration:
- M-25-21: Accelerating Federal Use of AI through Innovation, Governance, and Public Trust
- M-25-22: Driving Efficient Acquisition of Artificial Intelligence in Government
Supersede: Biden-era OMB M-24-10 (2024-03 federal AI use) and M-24-18 (2024-10 AI acquisition).
M-25-21 highlights: federal AI use
Section titled “M-25-21 highlights: federal AI use”Core pivot
Section titled “Core pivot”From “risk-prevention first” → “innovation-acceleration first”:
| Dimension | M-24-10 (Biden) | M-25-21 (Trump) |
|---|---|---|
| Framing | Safety / civil-rights protection | Innovation / public-service efficiency |
| Risk classification | ”Rights-impacting AI,” “safety-impacting AI” | Simplified “high-impact AI” |
| Chief AI Officer (CAIO) | Mandatory plus strong obligations | Retained but responsibilities reshaped |
| Public engagement | Mandatory public consultation | Weakened |
| Transparency | Annual AI use inventory plus detailed disclosure | Simplified disclosure |
| Risk management | NIST AI RMF as de facto baseline | NIST AI RMF still the baseline, but reduced documentation burden |
Retained obligations
Section titled “Retained obligations”- CAIO role: federal agencies continue to appoint a Chief AI Officer
- Annual public AI use inventory (simplified)
- High-impact AI risk assessment (with lighter documentation)
- Follow NIST AI RMF + GenAI Profile as the baseline
M-25-22 highlights: federal AI procurement
Section titled “M-25-22 highlights: federal AI procurement”Core pivot
Section titled “Core pivot”From “responsible procurement + vendor disclosure” → “efficient procurement + American-made AI preference”:
- Simplified vendor disclosure lists
- “Buy American AI” preference (U.S.-developed AI favored)
- Accelerated procurement timelines (up to 6-month cycles)
- LLM contracts: neutrality requirements aligned with the “Preventing Woke AI” EO
Expected impact
Section titled “Expected impact”- U.S. AI vendors (Anthropic, OpenAI, Google DeepMind, xAI, Meta, Microsoft) benefit
- Chinese vendors (DeepSeek, Doubao, Coze, etc.) effectively barred from federal procurement
- EU models (Mistral, etc.) face restricted access
Compliance implications
Section titled “Compliance implications”For LLM vendors
Section titled “For LLM vendors”- LLMs on federal contracts must pass the “Unbiased AI Principles” test
- Anthropic, OpenAI, and others must ready “Gov” variants (similar to existing Anthropic Gov / OpenAI Gov)
- Decommissioning costs clauses place replacement responsibility on the vendor when non-compliance arises
For federal agencies
Section titled “For federal agencies”- NIST AI RMF remains the de facto baseline
- Reduced compliance documentation burden
- But expanded AI adoption → greater operational pressure on data security and model evaluation
Relationship to the EO framework
Section titled “Relationship to the EO framework”| Layer | Instrument |
|---|---|
| Policy direction | EO 14179 (2025-01) |
| Strategy document | Trump AI Action Plan (2025-07) |
| Federal agency implementation guidance | OMB M-25-21 / M-25-22 (this page) |
| State-law preemption | EO 14365 (2025-12) |
Text and links
Section titled “Text and links”| Source | Link |
|---|---|
| OMB official memoranda | whitehouse.gov/omb |
| M-25-21 full-text PDF | download from OMB site |
| M-25-22 full-text PDF | download from OMB site |
| Industry commentary | Covington, Sidley, Paul Hastings analyses |
Version history
Section titled “Version history”| Date | Event |
|---|---|
| 2024-03-28 | Biden M-24-10 released |
| 2024-10-24 | Biden M-24-18 released |
| 2025-04-03 | Trump M-25-21 / M-25-22 released, superseding M-24-10/18 |